Vehicle Lane Change Intention Recognition And Lane Change Trajectory Prediction

Lane changing increases the risk of traffic accidents.Based on the recognition of vehicle lane change intention and the prediction of future trajectory,automatic driving can reduce the number of lane changes and reduce risks by calculating the safe distance with surrounding vehicles according to the predicted trajectory.Therefore,accurate trajectory prediction model plays an important role in improving the safety of lane change behavior.Based on NGSIM,which is the real vehicle track data,this paper identifies the vehicle lane changing intention and predicts its lane changing trajectory.In this paper,a variety of machine learning models are used to model multi-variable trajectory sequence of vehicles to realize lane change intention recognition,and ROC curve is used to compare the recognition accuracy.Long short-term Memory(LSTM)model was used to predict trajectory sequences,and Root Mean Squared Error(RMSE)was used to measure the sequence prediction results.Experiments show that increasing the characteristic parameters of surrounding vehicles can effectively improve the accuracy of intention recognition and reduce the error of trajectory prediction.The main research contents of this paper are as follows:In the first part,based on different rules,data extraction,data filtering,data filtering,data integration,data feature selection and other pre-processing steps are carried out for the lane-changing and straight track of different vehicles in the public data set NGSIM.Then,lane-changing or straight track data of different vehicles are extracted from the database using the method based on specific rules.The second part analyzes the lane changing behavior of expressway vehicles by using the track data of different vehicles extracted in the first step and identifies the lane changing intention of current vehicles at the moment through multiple variables of the current moment.In simulation experiment using experimental model of hidden markov model,the support vector machine(SVM)model,a decision tree model,a control model,the vehicle's trajectory sequence modeling,multivariate respectively using extract good normalization trajectory sequence data offline training,judgment or continue straight vehicle flow,model training accuracy reached 96.5%,respectively,99.7%,89.6%.The third part uses the long and short memory neural network trajectory sequence prediction model to train and test the historical trajectory samples of the current vehicle.According to the current and historical vehicle running status,the position,speed and other motion states of the vehicle in the predicted period are predicted,and the position of the vehicle at the next moment is predicted.The model was optimized by changing the length of the input and output time series of the model,the structure of the model,the characteristic parameters of the input and the influence of the surrounding vehicles.The root mean square error(RMSE)is used to measure the accuracy of trajectory prediction,and the results show that increasing the characteristics of trajectory input and appropriate time step can help the model converge better and faster,and output more accurate trajectory.To sum up,this paper carries out work on lane changing behavior recognition and trajectory prediction of vehicles on expressways,mainly including: multi-variable trajectory sequence modeling of vehicles by using a variety of machine learning models to realize lane changing intention recognition,and constructing trajectory sequence prediction model by using long and short memory neural network.